Intel has struggled in recent years, especially after the launch of the iPhone with Apple chips and the rise of an AI and data center chip industry largely dominated by Nvidia and AMD.
However, the company is not letting these developments pass by and has announced the introduction of its Xeon 6 chips and new Gaudi 3 AI accelerators.
The Intel Xeon 6900 P-core series, as the chips are called, offers up to 128 cores for extremely intensive AI workloads, doubling the performance of its predecessors, with higher core counts, more memory bandwidth, and built-in AI acceleration.
Power reinforcement
The new Gaudi 3 AI Accelerator chips, aimed at generative AI, feature 64 Tensor CPU cores, eight matrix multiplication engines, and 128GB of HBM2e memory. They deliver up to 20% more throughput at twice the price, and outperform Nvidia’s H100 chips for LLaMa 2 70B inference.
The specifications that Intel presents for the Gaudi 3 and Xeon 6 chips are downright impressive and would indicate that the company has managed to find an innovative way out of the impasse it finds itself in.
In addition to designing and manufacturing the chips, Intel has also partnered with Dell and Supermicro to work on “co-engineered systems” tailored specifically for those companies’ specific AI needs. Dell, for its part, is co-engineering RAG-based solutions that leverage Gaudi 3 and Xeon 6.
Pay attention to AMD – or not?
AMD has benefited the most from Intel’s largesse in the AI chip space, designing powerful chips that are then manufactured by TSMC. It’s a sound business model, and one that Intel has largely struggled to replicate or disrupt.
The company plans to release its fifth-generation EPYC Turin 3nm datacenter chips in the second half of 2024, based on its Zen 5 architecture and featuring up to 192 cores and 384 threads, more than on par with the Xeon and Gaudi chips, at least on paper.
In the closely followed Top500 listThe Intel-powered Aurora supercomputer, which is not yet fully operational, finished second behind the AMD-powered Frontier but did take first place in an AI benchmark.